Sorry. Looks like the code is gone. It could be patched up with some clever grep + bash-fu though

grep -Poh '(?<=^\(ns )[a-z_\.]+$' **/*.clj will get you a list of namespaces

depending on lein, clj, boot, you'll have to differ what you do, but something like: while read -r namespace; do clj -e "(require '${namespace})"; done And then you pipe these together. So

grep -Poh '(?<=^\(ns )[a-z_\.]+$' **/*.clj | while read -r namespace; do clj -e "(require '${namespace})"; done

I think it's the right stance until a solution is found which knows about the env.

actually, I do have an idea

you have the full list of namespaces from the initial analysis right? So if there's a use of a namespace from that list without an explicit require then it's a lint line.

There are still things like Java classes and goog/...

Of course we could whitelist goog

that doesn't matter though 🙂

you're blacklisting the namespaces you found in analysis.

(unless they are required)

you mean from the export?

analysis cache.

those two terms have too much overlap 🙂 too much ambiguity.

it could be a custom linter which works on the export if you want to do it like that. but clj-kondo should always be able to work without the initial cache as well, so it's not an option to use that for built-in linting

ah, but it will still work! 🙂

you just won't get that kind of lint.

if kondo doesn't know about any namespaces, then none of them will be on the blacklist, hence they will all be okay.

loading the entire cache isn't an option for built-in linting: it's too slow

I don't understand. Isn't the cache used for e.g. arity checks?

yes, but kondo carefully only loads what it needs from the cache

so, potentially, this could be carefully loaded?

it only loads namespaces from the cache which have been required by the user. it seems you want to know about all namespaces, even if they aren't required?

I have the same kind of problem with the private var linter. private vars might be validly used in other namespaces (e.g. foo.cljs + foo.cljc), but I can't really check for that while linting only one file. using the cache for this is too slow, because I'd have to scan all the previously linted namespaces as well

we only need to know if clojure.string exists or not, so the presence of the file is sufficient. The file doesn't need to be parsed necessarily. Alternatively you could spit out a tiny extra file with all the ns' in (assuming loading that doesn't take a long time that is)

if there is no cache yet, how can clj-kondo know about the existence of a namespace? it should be able to work with that

to prevent false positives

that's why you blacklist, not whitelist 🙂

(when (and (file/exists? "cache/clojure.string") (not (required? ana "clojure.string")) (add-lint-error!))

if there's no cache then file/exists? will not pass, and there will be no false positive.

scenario without cache: user types (foo/bar). clj-kondo doesn't know foo. so no error. scenario with cache: user types (foo/bar) and foo is in the cache. but hasn't been required. now you get an error?

that could work yeah

I'm finally getting it, thanks 😉

bingo!

and this will be performant. yeah

seems like a very good compromise imo 🙂

https://github.com/borkdude/clj-kondo/issues/515

better way to list all namespaces: echo "(require '[clojure.tools.namespace.find :as f]) (run! println (f/find-namespaces (map *command-line-args*)))" | clj -Sdeps '{:deps {org.clojure/tools.namespace {:mvn/version "0.3.1"}}}' - src